Pedro/add gemini 3 flash (#4323)
This commit is contained in:
@@ -436,6 +436,7 @@ class Settings(BaseSettings):
|
|||||||
"label": "Gemini 2.5 Flash",
|
"label": "Gemini 2.5 Flash",
|
||||||
},
|
},
|
||||||
"gemini-3-pro-preview": {"llm_key": "VERTEX_GEMINI_3.0_PRO", "label": "Gemini 3 Pro"},
|
"gemini-3-pro-preview": {"llm_key": "VERTEX_GEMINI_3.0_PRO", "label": "Gemini 3 Pro"},
|
||||||
|
"gemini-3.0-flash": {"llm_key": "VERTEX_GEMINI_3.0_FLASH", "label": "Gemini 3 Flash"},
|
||||||
"gemini-2.5-flash-lite": {
|
"gemini-2.5-flash-lite": {
|
||||||
"llm_key": "VERTEX_GEMINI_2.5_FLASH_LITE",
|
"llm_key": "VERTEX_GEMINI_2.5_FLASH_LITE",
|
||||||
"label": "Gemini 2.5 Flash Lite",
|
"label": "Gemini 2.5 Flash Lite",
|
||||||
@@ -474,6 +475,7 @@ class Settings(BaseSettings):
|
|||||||
"label": "Gemini 2.5 Flash",
|
"label": "Gemini 2.5 Flash",
|
||||||
},
|
},
|
||||||
"gemini-3-pro-preview": {"llm_key": "VERTEX_GEMINI_3.0_PRO", "label": "Gemini 3 Pro"},
|
"gemini-3-pro-preview": {"llm_key": "VERTEX_GEMINI_3.0_PRO", "label": "Gemini 3 Pro"},
|
||||||
|
"gemini-3.0-flash": {"llm_key": "VERTEX_GEMINI_3.0_FLASH", "label": "Gemini 3 Flash"},
|
||||||
"gemini-2.5-flash-lite": {
|
"gemini-2.5-flash-lite": {
|
||||||
"llm_key": "VERTEX_GEMINI_2.5_FLASH_LITE",
|
"llm_key": "VERTEX_GEMINI_2.5_FLASH_LITE",
|
||||||
"label": "Gemini 2.5 Flash Lite",
|
"label": "Gemini 2.5 Flash Lite",
|
||||||
|
|||||||
@@ -1292,10 +1292,23 @@ if settings.ENABLE_VERTEX_AI:
|
|||||||
litellm_params=LiteLLMParams(
|
litellm_params=LiteLLMParams(
|
||||||
api_base=f"{api_base}/gemini-3-pro-preview" if api_base else None,
|
api_base=f"{api_base}/gemini-3-pro-preview" if api_base else None,
|
||||||
vertex_location=settings.VERTEX_LOCATION,
|
vertex_location=settings.VERTEX_LOCATION,
|
||||||
thinking={
|
thinking_level="medium" if settings.GEMINI_INCLUDE_THOUGHT else "minimal",
|
||||||
"budget_tokens": settings.GEMINI_THINKING_BUDGET,
|
vertex_credentials=settings.VERTEX_CREDENTIALS,
|
||||||
"type": "enabled" if settings.GEMINI_INCLUDE_THOUGHT else None,
|
),
|
||||||
},
|
),
|
||||||
|
)
|
||||||
|
LLMConfigRegistry.register_config(
|
||||||
|
"VERTEX_GEMINI_3.0_FLASH",
|
||||||
|
LLMConfig(
|
||||||
|
"vertex_ai/gemini-3-flash-preview",
|
||||||
|
[],
|
||||||
|
supports_vision=True,
|
||||||
|
add_assistant_prefix=False,
|
||||||
|
max_completion_tokens=65536,
|
||||||
|
litellm_params=LiteLLMParams(
|
||||||
|
api_base=f"{api_base}/gemini-3-flash-preview" if api_base else None,
|
||||||
|
vertex_location=settings.VERTEX_LOCATION,
|
||||||
|
thinking_level="medium" if settings.GEMINI_INCLUDE_THOUGHT else "minimal",
|
||||||
vertex_credentials=settings.VERTEX_CREDENTIALS,
|
vertex_credentials=settings.VERTEX_CREDENTIALS,
|
||||||
),
|
),
|
||||||
),
|
),
|
||||||
|
|||||||
@@ -12,6 +12,7 @@ class LiteLLMParams(TypedDict, total=False):
|
|||||||
vertex_credentials: str | None
|
vertex_credentials: str | None
|
||||||
vertex_location: str | None
|
vertex_location: str | None
|
||||||
thinking: dict[str, Any] | None
|
thinking: dict[str, Any] | None
|
||||||
|
thinking_level: str | None
|
||||||
service_tier: str | None
|
service_tier: str | None
|
||||||
timeout: float | None
|
timeout: float | None
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user